Alzheimer disease (AD) is neuropsychiatric illness caused by multiple etiologies. Prediction of whether AD is genetically based in a given family is problematic because of censoring bias among unaffected relatives as a consequence of the late onset of the disorder, diagnostic uncertainties, heterogeneity, and limited information in a single family. We have developed a method based on Bayesian probability to compute values for a continuous variable that ranks AD families as having a major gene form of AD (MGAD). In addition, we have compared the Bayesian method with a maximum-likelihood approach. These methods incorporate sex- and age-adjusted risk estimates and allow for phenocopies and familial clustering of age at onset. Agreement is high between the two approaches for ranking families as MGAD (Spearman rank [r] = .92). When either method is used, the numerical outcomes are sensitive to assumptions of the gene frequency and cumulative incidence of the disease in the population. Consequently, risk estimates should be used cautiously for counseling purposes; however, there are numerous valid applications of these procedures in genetic and epidemiological studies.